An Independent Test Procedure for Heavy Goods Vehicle Blind Spot Safety devicesGRSG VRU-Proxi Bast 18th/19th July 2017

What is the problem that needs to be solved?

• Unseen pedestrians & cyclists injured by an HGV manoeuvring at low speed - ‘blind spot’ collisions. Two main crash mechanisms identified:• VRU (mainly pedestrians)

killed at front of HGV when it moves off from rest.

• VRU (mainly cyclists) killed at the side of an HGV when it turns left (right in EU)

• About equally important• Small number of collisions

at offside turning right

Research ProgrammesDirect Vision Protocol development Sponsored by TfL

Contractor: TRL/AVS

Blind Spot Safety systems Sponsored by TfL

Contractor: Thatcham Research, LowCVP, AVS, MIRA

Direct Vision Casualty impact Assessment Sponsored by TfL

Contractor: TRL/AVS

Collision & vehicle type Average number of fatalities per year

(2005-2014)Pedestrian Pedal

Cyclist

All collision types 82 15

Collision involving HGV>3.5t

11 5.3

Of which, involving HGV≥7.5t

9 4.7

HGV ≥7.5t ‘Moving off from rest’

3 1

HGV ≥ 7.5t ‘Turning left’ 1 3

Vulnerable Road User (VRU) Collisions In London

• Overall, pedestrians are a larger problem than cyclists

• Substantial annual variation in numbers (pedestrians 4-14; cyclists 2-9)

• Blind spots are a common contributory factor, particularly for moving off/ turning left

• Pedestrian and cyclist approximately equal in low speed manoeuvre

• Mis-coding may underestimate:• Tfl data suggests average 2 PC +

2 Ped fatalities/year in collision with construction HGV mis-coded as ‘other’

• +50%!

4% London traffic

21% of London Pedestrianfatalities [5-24%]

50% of London Pedal Cycle fatalities [26-60%]

The Disproportionate Involvement Of HGVs>7.5T and Construction Bodied HGVs

Note: Averages based on TfL data including miscoded ‘others’, which is only available for limited years. Ranges based on extremes from standard Stats19 (2005-15) and sample years for TfLs correction of miscoding

Collisions by Body type

• Both HGV and cycle move off from rest together, HGV Turns left

• Impact point typically nearside front• c.30% of Cyclist Left turn fatalities (n=18 in-depth

fatal cases)• ALL stopped at lights/give way before impact

Type 1

• Cyclist moves up inside of stationary HGV at speed. HGV moves off and turns left

• Impact point typically nearside front• c.40% of Cyclist Left turn fatalities• ALL stopped at lights/give way before impact

Type 2

• HGV and cycle both moving, sometimes cycle undertaking, sometimes HGV overtaking, low relative speed, HGV Turns left

• Impact point anywhere along full length• c.30% of Cyclist Left turn fatalities• None stopped at lights/give way before impact

Type 3

Dynamics of Left Turn CollisionsSource: Thatcham Research sponsored by TfL

Collision & vehicle type Average number of fatalities per year (2005-2014)

London Pedestrian GB Pedestrian

All collision types 82 519Collision involving HGV>3.5t

11 62

Of which, involving HGV≥7.5t

9 53

HGV ≥7.5t ‘Moving off from rest’

3 10

HGV ≥ 7.5t ‘Turning left’ 1 3

Application Outside London

• London sees 1/3 of relevant GB pedestrian collisions

Collision & vehicle type Average number of fatalities per year (2005-2014)

London Pedal Cyclist

GB Pedal Cyclist

All collision types 15 121Collision involving HGV>3.5t

5.3 22

Of which, involving HGV≥7.5t

4.7 18

HGV ≥7.5t ‘Moving off from rest’

1 2

HGV ≥ 7.5t ‘Turning left’ 3 7

Application Outside London

• London sees 1/2 of relevant GB pedal cyclist collisions

It happens everywhere, but London suffers more severely, and its not just related to high pedal cycle exposure

Application Across Europe

Source: CARE data in Knight (2011) assessing retrofit mirrors directive

• Clear evidence of the turning problem exists across EU

• c. 20% of HGV-VRU• Moving off from rest exists but

may be less prevalent than UK• c. 5% of HGV-VRU

• Study on revised Truck Front End Design (TFED) for EC DG Grow estimates:

• 278 fatal• 302 – 670 serious

• Pedestrian/Cyclist EU casualties in move off/turn left collisions

Source: CARE data (2015) combined with literature

Source: Volvo Trucks Safety Report 2017

‘Moving off’ not listed; likely under straight ahead?

UK left turn

Unclear which is representative for EU; shows similar outcomes occur on different infrastructure classes

UK (Robinson et al, 2016)

Mostly rigid vehicles Often stop at lights prior to collision Lateral expected to be small

Germany (Schreck & Seiniger, 2014)

Mostly articulated vehicles Rarely stop before collision Lateral separation can be 4m+

Application Across Europe: Differences in Left Turn Manoeuvres

Active & diverse aftermarket of

systems

Which has highest

casualty reduction potential?

Blind spot sensors: Why was an independent test procedure needed?

Categorising the systems in scopeField of View Aid•Technically, any system that helps enable a VRU in close proximity to be seen•However, direct vision and blind spot mirrors excluded because dealt with elsewhere

Proximity warning•System that uses sensors to detect the presence of a VRU close to the vehicle and warns the driver•Warning sounds whenever VRU is present irrespective of whether vehicles are on a collision path

Collision warning•System that uses sensors to assess the trajectories and speeds of both vehicles and warns when it

calculates a collision is imminent

Motion Inhibit•A system that prevents a vehicle moving off from rest when sensors detect a vulnerable road user in

close proximity to the front of the vehicle

Principles guiding design of tests• Realistic & Representative

• Road trials representative of normal driving but chances of encountering a collision scenario in short certification test is (fortunately!) very low

• Design off-road tests that closely represent typical collision scenarios

• Objective, Repeatable and Reproducible• Road trials are highly variable, so are humans• Use measurements, not judgements• Use robot controlled ‘dummies’ to simulate

humans and apply consistent steering

• Include consideration of HMI and False Positive• User feedback suggested high levels of driver

irritation with some systems leading to disabling of system

Test scenarios for each safety application

• Multiple tests and assessments undertaken

• Aim is to encourage• Good physical detection capability

• Ability to separate vulnerable road users from railings, signs etc

• Minimising false positives

• Employing good HMI to communicate warnings effectively to driver

• Not all tests applicable to all technologies

Note: HMI = Human Machine Interface

Test set up: Moving off from rest

With Vehicle Stationary aim to detect and warn in response to Adult crossing 0.3m in front

Child [3.5m] in front

Do not respond to Adult [3.5m] in front

Any clutter (railings, post hoardings etc)

Pedestrian until it moves in direction of vehicle

With Pedestrian stationary in front of vehicle attempt to move forward Start inhibit marks if cannot move (over-

rideable)

Test Set-up: Left turn

Detection in presence of clutter: HGV stationary 0.6m-1.5m away from kerb edged with metal railing. No warning until cyclist moves up inside and then warn

True Positives: Left turn manoeuvres when moving off together and when cyclist approaching from rear lateral separation 0.6m-1.5m Test vehicle speed 10 km/h

VRU speed 6-18 km/h

False positive: No warning when vehicle only passes, no turn, or where vehicle turns but no VRU (i.e. do not respond to kerbside clutter/pedestrian)

Start Position Impact Point

Start Position Impact Point

HMI Principles Adapted from a range of

existing automotive standards (e.g. ISO etc). Should be:

• Noticeable in the driving environment

• Distinguishable from other messages

• Indicative of direction of hazard

• Proportional to urgency of hazard

• Timely• Low nuisance level• Accepted by users

How does the driver react?

Example HMI Criteria

Evaluation Points Available

Result criteria

Proximity warning is issued over a single mode only (visual, audible or haptic).

3

0

Single mode

Multi-mode

The warning mode for proximity is visual or speech

1

0

Visual or speech

Tonal or Haptic

Visual proximity warnings are located within 15 cm of the upper, lower or forward facing edges of the passenger door window forward of the centre of the drivers seat base in its mid-point adjustment, without causing a visual obstruction to direct or indirect vision

1

0

In zone

Out of zone

Visual proximity warnings are amber in colour

1

0

Amber

Other colour

Speech warnings specify location of VRU (front, left side, right side)

1

0

Location specified

Not specified

Speech warnings comprise less than 6 words and take less than 2 seconds to complete

1

0

<6 words and 2s

6+ words or 2+s

Collision warnings are issued over more than 1 mode

0

1

Single mode [0]

Multi-mode [1]

Collision warnings are issued over at least 1 of Audible (Tonal) and Haptic modes.

1

0

Includes tonal and/or Haptic [1]

Does not include tonal and/or Haptic [0]

Visual collision warnings are located within 15 cm of the upper, lower or forward facing edges of the passenger door window forward of the centre of the drivers seat base in its mid-point adjustment, without causing a visual obstruction to direct or indirect vision

1

0

In zone [1]

Out of zone [0]

Visual collision warnings are red in colour

1

0

Red [1]

Other colour [0]

Audible tonal warnings have a signal to ambient ratio of specific loudness spectra greater than 1.3

1

0

≥1.3 [1]

< 1.3 [0]

Max points available 12 Total score

Total Score/Total Points Available

Scoring and weighting• Based mainly on

hypothetical systems at this time (2 real systems)

• Distinguishes between current systems but assumes imminent market arrival of OEM systems with greater functions (e.g. Wabco/Mercedes)

• Challenging nature should be subject to review and benchmarking of available and near market systems

Star rating Rating boundaries Nominal systems at each rating

0 Stars ≤10% 2, 3, 4, 5 and 9

1 Star >10% and ≤30% 1, 7, 10, and 11

2 Stars >30% and ≤50% 6 and 8

3 Stars >50% and ≤70% 14

4 Stars >70% and ≤90% 12

5 Stars >90% 13

Roadmap of future technologies

Proximity warning

Collision warning

Motion inhibit

AEB Pedestrian crossing

AEB City Turn Assist

In scope of current draft

Easily added in future update

TFL London Best practice rating covering

aftermarket progressing to OEM Vehicle starting from rest, limited

variation in pre-crash speed Low lateral separation (<1.5m) Discourage response to

inanimate objects & VRUs stationary on kerb

Encouraging best practice HMI/Warning

Bast Germany Regulatory Minimum standard

for OEM fit Vehicle always moving, wider

range of pre-crash speed High lateral separation (<4.5m) Prohibit response to stationary

objects unless they are VRUs Little control on HMI/Warning

Comparing proposalsProposals are procedurally similar and much in common but….

Conclusions Experience with systems in London suggests

High risk of ‘false positives’ with simple systems even when sensitive to only <1.5m lateral separation Results in ignoring warnings and/or disabling system via vandalism if necessary

If systems are required to be sensitive to 4.5m lateral then in London they will detect inside shops in many cases!

To ensure effectiveness either or both of following may be required Better avoidance of ‘false positive’ including stationary pedestrian on kerb

Correlation of warning urgency/intrusiveness with level of threat presented (e.g. amber visual proximity warning, red audio-visual collision warning

Need to ensure range of differential speeds and positions fully covers the ‘type 2 crash’ where HGV moves off from rest at lights while approached at speed from behind by pedal cycle

In GB a significant minority of VRUs are killed in front of an HGV as it moves off from rest Approximately equal in scale to cyclists in turns (maybe not across all EU?)

Could scope be expanded to include a ‘moving off’ function?